Near-channel classifier: symbiotic communication and classification in high-dimensional space

Area of Science:

• Brain-inspired computing
• Wireless communication systems
• Signal processing

Background:

• High-dimensional (HD) computing uses long, random vectors for robust data representation, especially at low signal-to-noise ratios (SNR).
• Hyperdimensional modulation (HDM) leverages HD representations for reliable wireless transmission, offering SNR gains.
• Existing HDM methods can be complex and may not fully integrate communication with classification tasks.

Purpose of the Study:

• To improve the efficiency and performance of hyperdimensional modulation (HDM).
• To develop a unified framework for near-channel classification (NCC) in wireless systems.
• To demonstrate the application of these methods in wearable sensor data processing.

Main Methods:

• Proposed methods to reduce HDM encoding/decoding complexity using bipolar or integer vectors instead of complex ones.
• Introduced a soft-feedback decoder to increase SNR gain and additive superposition capacity of HD vectors.
• Developed a near-channel classification (NCC) approach by combining communication and classification within a single framework using multifaceted HD representations.

Main Results:

• Achieved a 0.2 dB SNR gain with the soft-feedback decoder and increased additive superposition capacity up to 1.7 in noise-free conditions.
• Demonstrated NCC maintaining 94% accuracy for 5-class wearable gesture recognition at 0 dB SNR with 10,000-bit vectors.
• Reduced vector dimensionality to 2048 bits, showing graceful degradation with <6% accuracy loss at -5 dB SNR and interference from 6 nodes.
• Improved mean-squared error by up to 20 dB in reconstruction mode compared to standard decoding for 2048-dimensional vectors.

Conclusions:

• The proposed enhancements to HDM improve efficiency and robustness for wireless communication.
• The NCC approach reduces latency and complexity in distributed systems while maintaining high classification accuracy in noisy environments.
• These methods are effective for real-world applications like wearable gesture recognition using EMG sensors.